股票市场常被视为经济的晴雨表，是金融系统至关重要的组成部分。无论是不同国家或地区股市间，还是一个国家股市内部各子市场间，都存在着复杂多样的联系。准确描述和掌握它们间的相依关系，一方面有利于投资者理解市场结构，制定合理的投资策略。另一方面也有利于市场监管者控制市场风险和制定市场政策，具有重要的理论和实际意义。许多研究显示股票市场是具有非线性特征的复杂系统，而熵理论为研究非线性时间序列提供了有力的工具。因此本文采用熵理论结合核密度估计、网络理论作为主要研究方法，从信息角度探究股票市场间的相依关系。 首先，本文研究了国际股票市场间的相依关系。为避免数据离散化和数据长度有限带来的不利影响，将股市收益看做连续变量，采用核密度估计结合随机重排方法计算30个国际股市间的互信息。并在此基础上构建相依网络及其最大生成树，结果显示：相同地理区域的市场间拥有更强的相依关系，在最大生成树上相互聚集在一起。欧洲地区股市的国际相依程度最高，亚洲地区股市的国际相依程度最低。法国股市与其它市场关联最为紧密，而中国大陆股市仅与香港市场存在较强相依，与其它国际市场联系较弱。 由于金融市场间的相依关系为金融传染效应提供了途径，本文提出一种基于互信息和Bootstrap的金融传染效应检验方法，该方法基于数据驱动，适用于线性及非线性环境。运用该方法考察了美国次贷危机期间，30个国际股市间的金融传染现象。结果显示：传染效应在相同地理区域的市场间也更强。欧洲地区的传染效应最强，这与欧洲在次贷危机后爆发了严重债务危机的情况相符合。亚洲地区的传染效应最弱。 然后，本文分析了中国A、B、H股市场在2015年股灾前后的相依关系。根据市场状态将样本划分为：平静期、牛市期、股灾期和恢复期。分别计算各阶段市场间的互信息，并釆用滑动窗口方法分析互信息的动态变化。结果显示：股灾前，三市场间相依关系平稳。股灾增强了它们间的相依关系，股灾后它们相依关系减弱。还发现A股和B股市场拥有更强的相依关系，股灾期间A股和B股市场相依程度的增加幅度也更大。从股灾前夕开始，沪深B股市场间相依关系的变化趋势要领先于沪深A股市场。此外，本文还采用样本熵和广义Hurst指数考察了三市场的有效性，发现H股市场的有效性最高。 最后，本文探究了中国A股市场行业板块在2015年股灾前后的相依关系。对样本同样划分为：平静期、牛市期、股灾期和恢复期。各阶段及滑动窗口中板块间互信息值表明：股灾同样加强了行业板块间的相依关系，股灾结束后，板块间相依关系减弱。可选消费和工业板块与其它板块相依程度较高，金融板块与其它板块的相依关系最弱。板块间相依关系的变化较平稳。 同时，本文还利用传递熵分析了各行业板块间的信息传递。发现在平静期，板块间信息传递微弱。在牛市期，信息传递增加，公用板块在此时期影响最大，是信息传递的枢纽节点。在股灾期，板块间信息传递达到最大，市场有效性最低。股灾后，信息传递减小，但仍高于平稳期，金融板块成为接收信息最多的板块。板块间信息传递受市场行情显著影响。 关键词：熵理论；股票市场；相依关系；核密度估计；最大生成树

 The stock market which is an essential part of the financial system is usually considered as the barometer of the economy. There are complex and diverse connections between stock markets. To accurately describe and understand their dependence is not only useful for investors to understand the market structure and make reasonable investment strategies, but also helpful for market regulators to control risk and make polices. It has important theoretical and practical significance. Many researches have discovered that stock markets are nonlinear complex systems. Entropy theory provides a powerful tool for nonlinear time series analysis. So this dissertation combines entropy theory, kernel density estimation and network theory to explore the dependence in stock markets. Firstly, this dissertation researches the dependence in international stock markets. To avoid the adverse influence brought by data discretization and limited data length, the stock return is considered as continuous variable. Kernel density estimation and random shuffling method are used to calculate mutual information. Then the dependence network and its maximum spanning tree are built based on mutual information. The results show that markets which are in the same geographical region have more dependence. European markets have the strongest average dependence. Asian markets are lest dependent. France shares the most dependence with others. China is only relatively strong dependent with HongKong and weakly dependent with other markets. Since the dependence provides channels for the financial contagion. This dissertation proposes a financial contagion test method based on mutual information and Bootstrap. It is data-driven and can be used in linear and nonlinear conditions. The contagion effect of the US subprime crisis is tested with this method. The results display that the contagion effect is also stronger among markets within the same region. Europe experiences the strongest contagion effect. This is in accord with the fact that Europe has been heavily impacted by the subprime crisis and then the European debt crisis erupts. Asia experiences the weakest contagion effect. Secondly, this dissertation analyzes the dependence between China A, B, H share markets around the crash of 2015. According to the market condition, the full sample is divided into four stages: the tranquil, bull, crash and recovery period. Mutual information between the markets is calculated. Rolling windows are used to analyze the dynamics of the dependence. The results suggest that before the crash, the market dependence is stable. The crash enhances the dependence. After the crash, market dependence decreases. It is also found that the A and B share market has stronger dependence. In the crash, the dependence increasement of A and B share markets is also larger. From the eve of the crash, the change trend of the dependence between Shanghai and Shenzhen B share markets is ahead of their A share markets. In addition, Sample entropy and general Hurst exponent are applied to analyze the market efficiency, the result suggests that H share market is more efficient. Lastly, this dissertation studies the dependence among the industrial sectors of A share markets. The sample is also divided into: the tranquil, bull, crash, recovery period. Mutual information in each stage and rolling window indicates that the crash also enhances the dependence among sectors. When the crash is over, the dependence decreases. Consumer Discretionary and Industry sectors have strong dependence with others. The Financials sector has the weakest dependence. The evolution of the sectoral dependence is relatively stable. And, this dissertation applies transfer entropy to analyze the information transfer among the sectors. The results show that the information transfer is weak in the tranquil period. In the bull period, information transfer increases, Utility sector is most influential and is the hub node. In the crash period, the information transfer is the strongest of the four stages. The market is most inefficient. After the crash, information transfer decreases but is still higher than the tranquil period. The Financials sector receives the most information. Sectoral information transfer is sensitive to market status. Keywords: entropy theory; stock market; dependence; kernel density estimation; maximum spanning tree